Perturbation analysis of Poisson processes (Q2448701)

Let \(\Phi\) be a Poisson process on a measurable space \((\mathbb{X}, \mathcal{X})\). Denote by \(\Pi_{\lambda}\) the distribution of a Poisson process with intensity measure \(\lambda\) (\(\lambda\) is a \(\sigma\)-finite measure on \((\mathbb{X}, \mathcal{X})\)). Let \(f(\Phi)\) be some measurable function of \(\Phi\). Under certain assumptions on \(f\), the authors of [\textit{I. Molchanov} and \textit{S. Zuyev}, Math. Oper. Res. 25, No. 3, 485--508 (2000; Zbl 1018.49022)] derived a variational formula for the finite measure \(\lambda\). In the present paper, the author extends the variational formula to the \(\sigma\)-finite measure \(\lambda\). The extension of the identity from finite to \(\sigma\)-finite measures is a non-trivial task. The approach is based on the recent Fock space representation in [\textit{G. Last} and \textit{M. D. Penrose}, Probab. Theory Relat. Fields 150, No. 3--4, 663--690 (2011; Zbl 1233.60026)]. The paper is organized as follows. In Section 2, the author introduces some basic notation and recalls facts about the Fock space representation and likelihood functions of Poisson processes. Section 3 uses an elementary but illustrative argument to prove a simple version of the variational formula from [Molchanov and Zuyev, loc. cit.]. In Section 4, the author proves the main result of the paper. Section 5 derives conditions on \(\lambda\) that are necessary for the variational identity to hold for all bounded functions \(f\). In some cases, these conditions are also sufficient. Section 6 gives general Margulis-Russo-type formulas for derivatives. The final Section 7 treats perturbations of the Lévy measure of a Lévy process in \({\mathbb{R}}^d\).